| Description |
Modeling the relationships between climate and crop growth is an important tool to study agriculture in the past and future. Agricultural suitability models, which estimate the probability of rain-fed cultivation in a particular location, provide an opportunity to predict agricultural conditions for multiple scenarios in space and time. This thesis presents a methodological approach to developing a simple agricultural suitability model with machine learning techniques and applying it to two scenarios. First, to predict the potential impacts of climate change, the model was used to project agricultural suitability at the end of the 21st century at the global scale. Second, the model was used to predict agricultural suitability in Range Creek Canyon, an important archeological site in east-central Utah, in the 10th -13th centuries. The agricultural suitability model was supplemented with a streamflow model of Range Creek to address the potential for irrigation. The predictive power of random forest and XGBoost were compared and random forest was found to build the stronger model in this application. Applied to the future, this model predicts a net increase of 13.8-28.2% of agriculturally suitable land by AD 2100. The broad patterns agree with past suitability models, with the largest gains of suitability distributed in the high-northern latitudes. Decreases in suitability are projected in some regions, particularly in the current most intensely cultivated regions including Midwestern America. A sensitivity analysis revealed that the main driver of these shifts was changing growing season length and intensity, while changing soil moisture had a limited effect. Applied to the past in Range Creek Canyon, the model predicted very low agricultural suitability, providing evidence that rain-fed agriculture in Range Creek Canyon from AD 900-1200 was nearly impossible. Modeled Range Creek streamflow predicted that mean streamflow from AD 900-1200 was 6.34% greater than modern streamflow. The area of maize fields that could be irrigated with this amount of streamflow was approximately 30.6 hectares. The agricultural suitability model applied to two scenarios differing in scales of time and space demonstrate how it can provide meaningful insights to a broad range of past and future scenarios. |
